Search

We applied the bag of visual words model for visual texture to a dataset of realistic powder micrograph images drawn from eight closely related particle size distributions. We found that image texture based powder classification performance saturates at 89 ± 3 % with 640 training images (80 images per class). This classification accuracy is comparable to classification using conventional segmentation-based particle size analysis. Furthermore, we found that particle size distributions obtained via watershed segmentation are generally not statistically equivalent to the ground truth particle size distributions, as quantified by the two-sample Kolmogorov-Smirnov test for distribution equivalence. We expect image texture classification methods to outperform particle size analysis for more challenging real-world powder classification tasks by capturing additional information about particle morphology and surface textures, which add complexity to the image segmentation task inherent in particle size distribution estimation. [ABSTRACT FROM AUTHOR]

Fatigue crack initiation sites (FCISs) are crucial for fatigue investigations in metallic compounds, since they are closely related to the fatigue failure mechanisms and inform methods against crack initiation. However, the identification of FCISs requires professional knowledge and can be demanding and time consuming. Computer vision techniques show excellent performance for image texture recognition tasks and have been applied to some material domains. In order to explore more possibilities and widen the applications of computer vision techniques on material domains, as well as reduce the labor and knowledge requirements for FCISs recognition tasks, a state-of-the-art object detector – deeply supervised object detector which can train models from scratch, was introduced. The models trained from limited data show reasonable capacity for recognizing FCISs. Results demonstrate that increasing training dataset size can improve the accuracies of the models, while raising the number of epochs can result in a superior ability to recognize delicate features. Moreover, most of the images that cannot be recognized for FCISs possess common characteristics, such as, poor image quality, unclear features, insufficient training data and etc. Based on the above findings, several proposals are made for the future work to improve the performance of the same or relevant studies, as a reference. [ABSTRACT FROM AUTHOR]